Time series studies of air pollution and mortality: is semi-automated model selection possible?

Time series regression models are specially suitable in epidemiology for evaluating short-term effects of time-varying exposures. The objectives of this paper are twofold: 1) to apply transfer function models for regression analysis of epidemiological time series; 2) to explore the potential of semi-automated or automated approaches for model construction. The ideas are illustrated by analysing data on the relationship between daily non accidental deaths and air pollution in the 20 US largest cities.